Temporal K-Core Retrieval (TKCR)

Official implementation of "Temporal K-Core Pruning: Enhancing Efficiency and Timeliness in Large-Scale Graph-Enhanced Retrieval" (WSDM 2026).

Overview

TKCR combines temporal filtering with k-core decomposition to achieve efficient and time-aware retrieval in dynamic knowledge graphs. Our method:

• Achieves 100% temporal precision while maintaining competitive MRR (0.802 vs 0.771 baseline)
• Reduces search space by 62.5%-86.5% depending on k value
• Outperforms SOTA methods with 28.0% higher MRR than ColBERT and 8.1% higher than HyDE
• Requires no model training or expensive LLM calls

Installation

# Clone the repository
git clone https://github.com/anonymous/temporal-kcore-retrieval.git
cd temporal-kcore-retrieval

# Create conda environment
conda create -n tkcr python=3.8
conda activate tkcr

# Install dependencies
pip install -r requirements.txt